Fabric Defect Detection for Industrial Use
The textile industry, like all other industries, has been faced with the need to rise to the occasion and meet the demands in these competitive times. Sizable demands put forth by ever-changing fashion trends and quality quotas can only be met by evolving technology. Looms have, in response, become faster and more precise in response to these requirements. However, the mechanism of work applied is one that cannot be freed from error. It is unavoidable that the resulting fabrics will have defects hidden within the textures. These defects, once identified, can be rectified through established means. The problem lies in spotting these defects in the first place. Over time, numerous automated systems have been suggested as solutions, of which a few are suited to actual industrial applications.
This paper walks through a few methods of tackling the problem. Unlike a survey paper, the aim is not to briefly introduce or categorize attempted approaches but rather to provide more substantial insight into possible viable implementations. The implementations discussed are centered on (1) Yarn Tracking with the use of a CNN, (2) Fourier Transforms on the images to spot non-uniformity and (3) Image Reconstruction to create a reference texture and subtract the sample image respectively.
Keywords: Fabric Defect Detection, Image Processing, Fourier Transforms, Deep Learning, Convolutional Neural Networks, Yarn Tracking.